🧠 Build Your Own Python Coding Agent (With Tool Use!)

Ever wanted to make your own agent ? Not just a chatbot, but something that can actually do things? How tools like Devin or AI-powered IDEs actually work? What if you could build one, an agent that reads your code, makes changes, and creates new files based on your instructions?

This repo is a tiny, tinkerable playground to explore that idea. I'll walks you through how to build a language model powered agent that can read, write, and modify files in your project folder using only Python and a model like LLaMA 3.

---

🧭 What Are We Exploring Here?

You’re not building an app. You’re building an autonomous coder. YES AN AUTONOMOUS CODER sounds like a lot right.

One that:

• Understands your request ("Summarize this file" or "Create a new script with X").
• Thinks about what tools it needs.
• Uses those tools to take action like writing to a file or listing code in a directory.

This is less about making a coding agent SAAS product and more about learning how agents think and act.

---

📦 What You Get

Out of the box, this agent can:

✅ Read files
✅ Create new files
✅ List files in a folder
✅ Append or overwrite existing code
✅ Use LLaMA 3 via OpenRouter to decide what to do

All orchestrated through a few small Python scripts.

---

🚀 Getting Started

1. Clone this repo

git clone <your-repo-url>
cd <your-folder>

2. Install dependencies

pip install -r requirements.txt

This uses:

openai
python-dotenv

3. Add your model API key

Create a .env file with:

MODEL_API_KEY=your-openrouter-api-key

You can grab a key at https://openrouter.ai.
I'm using meta-llama/llama-3.3-8b-instruct:free because it's free, but you can swap this easily in main.py.

---

🧪 Try It Out

Run the agent:

python main.py

You’ll see:

Chat with LLAMA (type 'exit' or 'ctrl+c' to quit)

Now try talking to it:

Aryan👽: hey can you read tools.py?

AI 🤖: Sure! Here's a quick summary of tools.py:
- read_file: reads content of a file
- write_file: replaces file content
- list_files: lists files in a directory
...

Or:

Aryan👽: create a new file called test.py
AI 🤖: Done! Want me to add anything to it?

Or even:

Aryan👽: append a function to multiply two numbers to test.py
AI 🤖: Appended. Now the file contains both sum and multiply functions.

It’s magical and it’s simple.

---

🔧 The Tools

Here’s what the agent has in its toolbox:

Tool	What it does
read_file	Reads and returns file contents
write_file	Overwrites a file with new content
append_file	Adds content to the end of a file
create_file	Makes a new file
list_files	Lists all files recursively in a folder

Each tool is defined in tools.py, wrapped in a function + metadata (name, description, expected input). The agent chooses when to use them.

---

🧠 How It Works

At its core:

• You talk to it via CLI (cli.py)
• It uses a local agent (agent.py) to process your prompt
• The agent builds a conversation, decides when to call a tool, and sends all of that to the LLM
• If the LLM wants to call a tool, we do it then feed the result back to the model.
• Continue the conversation.
• Get the desired output.

It’s like giving your LLM a memory and a toolbox to work for you.

---

🪜 Example Walkthrough

You might start with:

Aryan👽: list all .py files
AI 🤖: ['cli.py', 'agent.py', 'main.py', 'tools.py']

Then say:

Aryan👽: what's inside agent.py?
AI 🤖: It defines the Agent class, handles tool usage, wraps messages, and communicates with the LLM.

Then try:

Aryan👽: write a 2 sum function into cc.py
AI 🤖: All set. You can now run cc.py to test the function.

You’re not copying and pasting code. You’re describing intent and the agent takes care of the rest.

---

📁 Folder Structure

.
├── agent.py        # Core agent logic
├── cli.py          # Command-line interface
├── main.py         # Entry point
├── tools.py        # All available tools
├── .env            # API key config
├── requirements.txt

---

✨ Extend It Yourself

Want to add your own tool? Super simple:

Define a function first for what the tool will do.

new_tool = Tool(
    name="your_tool",
    description="What this tool does and what input it expects.",
    func=your_function
)

Then just add it to the tools list in main.py. Boom — your agent has a new skill.

---

🌱 Why This Matters

LLMs are good at talking, but they become powerful when they can act.

This repo is a seed for building agents that don’t just answer questions but do things. Start here. Then scale up. Add memory. Add planning. Add real-world tasks. Build More.

---

📬 Reach Out

Curious how this works? Want to jam on agents, dev tools, or new ai trends in general ?

Feel free to fork, experiment, and reach out.

Happy building! 🤖